ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
Contents lists available at ScienceDirect
ISPRS Journal of Photogrammetry and Remote Sensing
journal homepage: www.elsevier.com/locate/isprsjprs
The potential of the greenness and radiation (GR) model to interpret
8-day gross primary production of vegetation
Chaoyang Wu a,b,⇑, Alemu Gonsamo b, Fangmin Zhang c, Jing M. Chen b
a
State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing, China
Department of Geography, University of Toronto, 100 St. George St., Toronto, ON M5S 3G3, Canada
c
Jiangsu Key Laboratory of Agricultural Meteorology, College of Applied Meteorology, Nanjing University of Information Science and Technology, Nanjing, Jiangsu 210044, China
b
a r t i c l e
i n f o
Article history:
Received 20 June 2012
Received in revised form 21 September
2013
Accepted 27 October 2013
Available online 20 December 2013
Keywords:
Enhanced vegetation index
Flux
Gross primary production
Remote sensing
Climate change
Carbon cycle
a b s t r a c t
Remote sensing of vegetation gross primary production (GPP) is an important step to analyze terrestrial
carbon (C) cycles in response to changing climate. The availability of global networks of C flux measurements provides a valuable opportunity to develop remote sensing based GPP algorithms and test their
performances across diverse regions and plant functional types (PFTs). Using 70 global C flux measurements including 24 non-forest (NF), 17 deciduous forest (DF) and 29 evergreen forest (EF), we present
the evaluation of an upscaled remote sensing based greenness and radiation (GR) model for GPP estimation. This model is developed using enhanced vegetation index (EVI) and land surface temperature (LST)
from the Moderate Resolution Imaging Spectroradiometer (MODIS) and global course resolution radiation data from the National Center for Environmental Prediction (NCEP). Model calibration was achieved
using statistical parameters of both EVI and LST fitted for different PFTs. Our results indicate that compared to the standard MODIS GPP product, the calibrated GR model improved the GPP accuracy by reducing the root mean square errors (RMSE) by 16%, 30% and 11% for the NF, DF and EF sites, respectively. The
standard MODIS and GR model intercomparisons at individual sites for GPP estimation also showed that
GR model performs better in terms of model accuracy and stability. This evaluation demonstrates the
potential use of the GR model in capturing short-term GPP variations in areas lacking ground measurements for most of vegetated ecosystems globally.
Ó 2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier
B.V. All rights reserved.
1. Introduction
Gross primary production (GPP) represents the gross carbon (C)
fixed by terrestrial ecosystems through photosynthesis, which in
turn drives several ecosystem functions (Beer et al., 2010). Therefore, accurate estimation of GPP can be a great advantage for quantifying net ecosystem production (NEP) and thus of particular
importance for global C cycle research (Zhao and Running, 2010).
Remote sensing technique offers a unique tool for monitoring
land surface globally and timely and hence several algorithms
are proposed to estimate GPP based on a combination of remote
sensing observations and ground measurements (Field et al.,
1998; Zhao et al., 2005, 2006; Yuan et al., 2010; Wu et al., 2009).
The main concept for estimating GPP from remote sensing observations is based on the Monteith (1972) logic,
GPP ¼ LUE APAR
ð1Þ
⇑ Corresponding author at: Department of Geography, University of Toronto, 100
St. George St., Toronto, ON M5S 3G3, Canada. Tel.: +1 647 524 0310.
E-mail address: [email protected] (C. Wu).
where LUE represents the light use efficiency, and APAR is the absorbed photosynthetically active radiation calculated as the product
of an absorbed fraction (fAPAR) and the amount of incident photosynthetically active radiation (PAR).
Following Monteith’s equation, GPP can be determined by estimating LUE and APAR separately and such method has shown
promising results in designing satellite-based GPP models (e.g.,
the Moderate Resolution Imaging Spectroradiometer (MODIS)
GPP products, Zhao et al., 2005; and the Vegetation Photosynthesis
Model (VPM), Xiao et al., 2005). Vegetation indices (VIs), which
provide measures of surface canopy greenness and vegetation
status, are the most widely used metrics in driving remote sensing
GPP models. A number of VIs have been shown to be useful in
estimating GPP, including the normalized difference vegetation
index (NDVI) (Rouse et al., 1974), the enhanced vegetation index
(EVI) (Huete et al., 2002), the MERIS Terrestrial Chlorophyll Index
(MTCI) (Dash and Curran, 2004), the land surface water index (Xiao
et al., 2005) and the wide dynamic range vegetation index (WDRVI)
(Gitelson, 2004). These VIs are incorporated in many existing GPP
models, either independently (Gitelson et al., 2008; Harris and
Dash, 2010; Kalfas et al., 2011) or in combination with other
0924-2716/$ - see front matter Ó 2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.isprsjprs.2013.10.015
70
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
variables (e.g., temperature and evaporation) (Xiao et al., 2006;
Coops et al., 2007; Gitelson et al., 2006; Sims et al., 2008; Hilker
et al., 2008a; Zhang et al., 2009; Sakamoto et al., 2011; Wu et al.,
2011; Peng and Gitelson, 2012).
One of the main sources of uncertainties in remote sensing
based GPP models is the uncertainty in the LUE calculation, which
is often achieved by an empirically determined maximum LUE for a
specific land cover type. Climate variables are used to modulate the
fixed values of land cover specific maximum LUE (Running et al.,
2004). However, the dependence on input climate variables and
the fact that each biome is assigned a single maximum LUE value
can cause significant deviation of estimated GPP from observations
(Zhao et al., 2006; Kanniah et al., 2009; Mu et al., 2011). In particular, spatiotemporal variations of LUE is a challenging issue for
such models and increasing evidence shows that misinterpreting
the spatiotemporal characteristics of LUE would significantly affect
the modeled GPP (Turner et al., 2003; Wang et al., 2010). To compensate the empirically determined LUE, specific indicators are
proposed to track LUE changes and thereby mitigating the dependence on input climate variables. For example, the photochemical
reflectance index (PRI) (Gamon et al., 1997) has been demonstrated
to have certain potential in estimating LUE. However, satellite
determination of LUE using PRI alone is confounded by the canopy
structure and the view observer geometry (Hilker et al., 2008b;
Hall et al., 2011).
Recent studies have turned to other alternative models instead
of the LUE concept for the estimation of GPP. For example, the temperature and greenness (TG) model proposed by Sims et al. (2008)
that combines MODIS EVI and LST has shown good potential in
modeling GPP for North America ecosystems. In particular, Gitelson et al. (2006) developed a chlorophyll content based GPP model
in crops that predicts GPP using only two input parameters, i.e.,
chlorophyll index and incoming PAR. Later analysis showed that
this model can provide better GPP estimates than that of the TG
model for several flux sites in North America, including both deciduous forest and evergreen forests (Wu et al., 2011). However, there
are two main difficulties for the application of such model for
operational use. The first is the radiation data that are used to
run the model. Most of the previous evaluations of this model used
radiation data from site level meteorological measurements (Wu
et al., 2010, 2011; Kalfas et al., 2011; Peng et al., 2011). Therefore,
it would be difficult to explicitly compare the model estimates
with the standard MODIS GPP product that uses coarse resolution
radiation data since evidence has shown that the uncertainty in
radiation will significantly affect the GPP outputs (Sakamoto
et al., 2011). A second difficulty is the feasibility of the model for
estimating and modeling short-term GPP variations in globally diverse vegetation ecosystems. Understandings of such uncertainties
are especially important for upscaling the algorithm to map regional and global GPP where ground observations are lacking. Here we
present the interpretation of short-term (8-day) GPP estimates
derived using remote sensing observations and global coarse
resolution radiation data (biases are generally within 50 W/m2
depending on regions) over 70 global C flux sites. These sites cover
various PFTs and thus provide an opportunity to test the robustness of the model and to further develop upscaling strategies.
Therefore, the objective of this study is the global calibration and
validation of the greenness and radiation (GR) model with respect
to different PFTs. Furthermore, we compare the performances of
the GR model and the standard MODIS GPP product against the
ground measurements.
2. Materials and methods
2.1. Study sites
We base our analysis on 70 representative C flux tower sites,
covering various vegetation ecosystems with latitudes ranging
from 40°S to 70°N and longitudes from 140°E to 130°W
(Fig. 1). The rules for site selection were mainly regulated by the
data availability (at least 2 years of continuous and complete data
records), data quality (less than 30% of gap-filled data), availability
of site-level meteorological data (radiation and temperature), and
no recent disturbances (fire, insect and harvest). To analyze the
influences of PFT on model simulations, we grouped the sites into
three PFTs: 24 non-forest vegetation (NF) sites, 17 deciduous forests (DF) sites and 29 evergreen forest (EF) sites. Detailed descriptions of each site and their regional climates are shown in
Supplementary Table 1.
2.2. Flux and meteorological data
The C flux data were downloaded from the following flux
networks, AmeriFlux (http://public.ornl.gov/ameriflux/dataproducts.shtml), Fluxnet-Canada (www.fluxnet-canada.ca), EuroFlux
(www.europe-fluxdata.eu), CarboAfrica (www.europe-fluxdata.
Fig. 1. Spatial distribution of the flux sites in this study, the NF ( ), DF ( ), and EF ( ) represent non-forest deciduous forest, and evergreen forest ecosystems, respectively.
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
71
eu/newtcdc2/CarboAfrica_home.aspx), and the Public FLUXNET
Dataset Information (www.fluxdata.org).
Since these data were acquired from different regional flux
networks, two gap-filling procedures were applied. For FluxnetCanada sites, the estimation of GPP and Re were achieved by empirical relationships: (1) nighttime NEE and nighttime temperature
and (2) daytime GPP and PAR. The procedures for gap-filling were
described in Barr et al. (2004) which was the standard method for
all Fluxnet-Canada sites. For sites from other regional flux networks, level-4 products were used which contain gap-filled and
u (typically value of u was 0.35 m s1 but may differ slightly for
individual sites) filtered records of C fluxes with quality flags for
both original and gap-filled data. The half-hourly measurements
were gap-filled using the Artificial Neural Network (ANN) method
Fig. 2. Relationships between shortwave solar radiation acquired from the National
Center for Environmental Prediction (NCEP) reanalysis II data and radiation of on
site measurements. The NF (s), DF (4), and EF (h) represent non-forest, deciduous
forest, and evergreen forest ecosystems, respectively.
Fig. 3. Relationships between flux-measured 8 day gross primary production (GPP)
and the product of enhanced vegetation index and photosynthetic radiation
(EVI PAR) for (a) non-forest (s), (b) deciduous forest (4), and (c) evergreen
forest (h), respectively.
72
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
(Papale and Valentini, 2003) and/or the Marginal Distribution Sampling (MDS) method (Reichstein et al., 2005). Although different
gap-filling methods were used, the reliability of multiple site comparisons and syntheses can be assumed as most methods tended to
cluster on similar results to within 10% (Desai et al., 2008), which
also agrees with the general understanding that biases associated
with different gap filling methods tend to be small and supports
our analysis across multiple sites (Papale et al., 2006). Apart from
the aggregated 8-day composite of GPP, we also calculated the 8day air temperature, global radiation and precipitation from the
site meteorological measurements.
2.3. MODIS products
We used three MODIS land surface products in this study,
which were acquired from the Oak Ridge National Laboratory’s
Distributed Active Archive Center (DAAC) website (http://daac.ornl.gov/MODIS/). The first product is the 8-day MODIS Terra Surface Reflectance Product (MOD09A1, 500 m, collection 5), which
provides surface reflectances for seven bands centered at 648 nm,
858 nm, 470 nm, 555 nm, 1240 nm, 1640 nm, and 2130 nm (Vermote et al., 1997). Pixels that are contaminated by cloud and aerosol were removed using the quality flag and only good data
were selected. Reflectance was extracted from 3 3 MODIS pixels
(1.5 km 1.5 km) that centered on the flux tower, which is similar
with previous analyses (Sims et al., 2008; Wu et al., 2011). Reflectances from three spectral bands, blue (Rblue), red (Rred), and nearinfrared (RNIR) were used to calculate the EVI as in Huete et al.
(2002):
EVI ¼ 2:5 RNIR RRed
1 þ RNIR þ 6 RRed 7:5 RBlue
ð2Þ
We have not use the standard MODIS EVI product because the
temporal resolution of that product is 16 days while our objective
is to interpret 8-day GPP.
We also extracted MODIS 8-day Land Surface Temperature
(LST) product (MOD11A2, 1 km, collection 5) derived by applying
the generalized split-window algorithm. In the split-window algorithm, emissivity in bands 31 and 32 are estimated from land cover
types, and atmospheric column water vapor and lower boundary
air surface temperature are separated into tractable sub-ranges
for optimal retrieval (Wan, 2008). The data quality was checked
through the quality flag such that only good data in the original
L1B data were used and LST contaminated by cloud effects and
poor retrievals were excluded. 3 3 Pixels around the flux site
were extracted to represent the flux tower (Wu et al., 2010).
The third MODIS product was the 8-day MODIS GPP product
(MOD17A2, 1 km, collection 5.1), which is included in this study
for model comparison. MODIS GPP is driven by daily MODIS land
cover, fAPAR, leaf area index (LAI) and interpolated surface meteorology at 1 km for the global vegetated land surface (Zhao et al.,
2006). This product is calculated as:
GPP ¼ emax mðT min Þ mðVPDÞ FPAR SWrad 0:45
ð3Þ
where emax is the maximum LUE obtained from lookup tables on the
basis of land cover type. The scalers m(Tmin) and m(VPD) reduce emax
under unfavorable conditions of low temperature and high VPD.
Tmin, VPD and SWrad are obtained from large spatial-scale meteorological data sets that are available from the NASA Global Modeling
and Assimilation Office (GMAO) (http://gmao.gsfc.nasa.gov/).
Observations with quality flag of ‘‘good’’ and ‘‘best’’ were used for
each site in 8-day interval and the mean value of the 3 3 pixels
around the flux site was extracted for the comparison (Wu et al.,
2012).
2.4. Global radiation data
To upscale the GR model and facilitate the model comparison
with standard MODIS GPP, we downloaded the daily downward
shortwave solar radiation flux from the National Center for
Table 1
Relationships between flux-measured GPP and EVI PAR for each individual site.
Site ID
Variables
Site ID
Variables
Non-forest
Slope
Intercept
R2
Deciduous forest
Slope
Intercept
R2
Evergreen forest
Site ID
Variables
Slope
Intercept
R2
US-ARM
US-NE3
DE-GEB
DE-KLI
DK-RIS
NL-LUT
UK-ESA
US-AUD
US-FPE
US-GOO
US-VAR
US-WKG
DE-MEH
DK-LVA
FR-LQ1
IL-AMP
IT-MAL
NL-CA1
NL-HOR
IT-PIA
CA-FEN
US-TON
Au-HOW
ZA-KRU
0.28
0.68
0.57
0.58
0.46
0.75
0.87
0.29
0.23
0.36
0.32
0.31
0.40
0.48
0.42
0.24
0.40
0.47
0.65
0.22
0.34
0.44
0.47
0.47
5.37
22.00
8.85
11.10
1.41
12.67
9.28
6.40
0.89
0.65
2.69
8.20
3.38
8.63
1.73
5.00
9.64
2.14
1.60
3.12
2.96
5.89
14.09
14.35
0.39
0.60
0.76
0.80
0.76
0.73
0.77
0.27
0.36
0.51
0.25
0.58
0.88
0.48
0.72
0.40
0.45
0.76
0.88
0.15
0.71
0.61
0.23
0.56
US-HA1
US-BAR
US-MMS
US-MOZ
US-SYV
US-WCR
BE-BRA
DE-HAI
DK-SOR
FR-FON
FR-HES
IT-COL
IT-NON
IT-PT1
IT-RO1
IT-RO2
CA-OAS
0.54
0.45
0.46
0.34
0.48
0.59
0.37
0.44
0.68
0.54
0.49
0.39
0.49
0.73
0.38
0.54
0.54
14.65
12.19
7.74
1.64
7.00
12.75
2.53
1.61
1.83
0.42
8.42
5.91
8.24
9.34
2.67
4.60
6.84
0.75
0.84
0.76
0.77
0.74
0.78
0.72
0.76
0.80
0.57
0.78
0.83
0.62
0.93
0.70
0.64
0.79
US-KS2
AU-TUM
FR-PUE
IT-LEC
PT-ESP
US-HO1
CA-NS3
US-ME2
US-NR1
CZ-BK1
DE-WET
FI-HYY
FI-SOD
FR-LBR
IT-BON
IT-LAV
IT-REN
IT-SRO
NL-LOO
SK-TAT
UK-GRI
CA-CA1
CA-CA3
CA-CA2
CA-MAN
CA-TP1
CA-TP3
CA-OBS
CA-OJP
0.28
0.54
0.20
0.16
0.26
0.56
0.39
0.56
0.38
0.60
0.68
0.55
0.43
0.39
0.34
0.50
0.45
0.44
0.51
0.28
0.42
0.59
0.37
0.40
0.49
0.22
0.47
0.55
0.45
16.43
5.97
13.10
13.55
17.32
11.53
3.90
4.30
8.68
1.32
3.83
1.82
0.91
10.90
13.74
8.57
1.53
11.22
9.72
2.11
5.57
7.96
3.72
5.97
3.54
1.87
9.15
4.80
5.75
0.44
0.66
0.42
0.21
0.20
0.78
0.55
0.74
0.48
0.68
0.67
0.78
0.75
0.65
0.56
0.70
0.57
0.54
0.83
0.76
0.50
0.83
0.76
0.69
0.61
0.72
0.72
0.70
0.62
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
Environmental Prediction (NCEP) reanalysis II data (http://
www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis2.html).
NCEP daily reanalysis II data provides the global radiation data divided into 192 94 Gaussian grids (1.875°) and this dataset has
been demonstrated to be appropriate for driving the MODIS global
GPP algorithm (Zhao and Running, 2010). This dataset is an improved version of the NCEP reanalysis I model that fixes errors
and updates parameterizations of physical processes and is thereby capable of capturing major changes in the surface climate
anomalies (Betts et al., 2006). Radiation for each site was determined using the geographic location (latitude and longitude) of
flux tower sites after bilinearly interpolating the NCEP data into
1 km grids.
73
26.0% and 38.1%, respectively, suggesting the largest variation in
the ability of EVI PAR to predict GPP for EF sites.
3.2. Model calibration and comparison with MODIS GPP
Modeling the slopes and intercepts of the relationships between
GPP and EVI PAR using remote sensing observations would
2.5. The GR model
We used a remote sensing based greenness and radiation (GR)
model for estimating GPP based on EVI and incoming PAR inputs
(Gitelson et al., 2006). The underlying mechanism of this model lies
in the correlation between the GPP and the total canopy chlorophyll content (Peng et al., 2011; Peng and Gitelson, 2011; Sakamoto et al., 2011). Therefore, vegetation indices that are proxies of
total chlorophyll content together with incoming PAR can be used
for GPP estimation. This GR model can be expressed as follows:
GPP ¼ SlopeðEVI PARÞ þ Intercept
ð4Þ
One of the uncertainties in the application of this model is the
calibration with diverse ecosystems to determine the Slope and
Intercept with respect to various PFTs. A previous effort in this
aspect has shown promising results at monthly time step across
North American flux sites (Wu et al., 2011). However, the feasibility of this model has not been demonstrated at shorter time scales
of various ecosystems globally. In this analysis, the Slope and
Intercept were determined by the following approach. We first
calculated the relationship between GPP and EVI PAR at each
individual site. Then slopes and intercepts for sites of the same
PFT were then empirically determined using site-level variables
for each PFT. For data of individual site, we used all observations
within a year (i.e., both leaf-on and leaf-off season) in later
analysis.
3. Results
3.1. Relationship between measured GPP and EVI PAR
We first compared the coarse resolution radiation data with
measured radiation data for each PFT (Fig. 2). Close relationships
were found between these two radiation datasets for each PFT
(R2 > 0.85, p < 0.001), assuring the use of the coarse resolution radiation dataset for both the standard MODIS GPP product and the GR
model.
We then explored the relationship between flux-measured GPP
and EVI PAR for each PFT (Fig. 3). GPP grouped into three PFTs
generally correlated well with EVI PAR with R2 of 0.54
(p < 0.001), 0.69 (p < 0.001) and 0.56 (p < 0.001) for NF, DF and EF
sites, respectively. The slopes of each PFT imply that EVI PAR
would overestimate GPP without model calibration. We have also
shown the potential of EVI PAR in predicting GPP for each individual site (Table 1). For a single site data, we observed substantial
differences in the R2 of GPP vs. EVI PAR relationship ranging from
0.15 (p < 0.001) for IT-PIA and 0.93 (p < 0.001) for IT-PT1. DF sites
had the best results (R2 > 0.70) and stable GPP and EVI PAR relationships, which was indicated by the lowest coefficient of variation (CV = 11.5%) in R2. The CV of R2 for NF and EF sites were
Fig. 4. Relationships between MODIS land surface temperature (LST) and the on
site measured air temperature. The NF (s), DF (4), and EF (h) represent non-forest,
deciduous forest, and evergreen forest ecosystems, respectively.
74
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
greatly favor its application across diverse ecosystems globally. To
achieve this, we focused on the estimation of the slopes and intercepts for each PFT using statistical parameters from EVI and LST,
both of which were shown to be useful in GPP model calibration
(Sims et al., 2008; Wu et al., 2011).
To support the inclusion of the satellite LST, we first compared
all LST with the flux-measured air temperature for each PFT
(Fig. 4). Generally, MODIS LST is a reasonable proxy of air temperature with R2 of 0.61 (p < 0.001), 0.52 (p < 0.001) and 0.61
(p < 0.001) for NF, DF and EF sites, respectively. MODIS LST tends
to be lower than air temperature in the lower LST range while
higher than air temperature in the high LST range. Such distribution has also been observed in previous study (Sims et al., 2008;
Wu et al., 2010) (note that some of the same data were used in
the present study, and thus these results are not fully
independent).
Based on previous results of Wu et al. (2011), we try to explain
the slopes and intercepts by a range of statistical variables of both
EVI and LST. Statistical parameters of EVI included the minimum
and maximum (EVImin and EVImax), the difference between
minimum and maximum (dEVI), the average (EVIave) and the standard deviation of EVI (EVIsd). Statistical parameters of LST were the
average and standard deviation of LST (LSTave and LSTsd). We found
that the slopes and intercepts for each PFT can be explained by the
single or combined forms of these statistical parameters (Fig. 5).
For the NF sites, the slopes of each site were found to be correlated
with dEVI (R2 = 0.52, p < 0.001) and the variance in intercepts can
be explained by dEVI LSTave (R2 = 0.47, p < 0.001). For the DF
sites, the slopes and intercepts were linear functions of EVIsd/dEVI
(R2 = 0.52, p = 0.001) and LSTave/EVIave (R2 = 0.53, p < 0.001). Similarly, significant relationships were identified between the slopes
and dEVI LSTave (R2 = 0.50, p < 0.001) and between the intercepts
and the LSTave/EVIsd (R2 = 0.60, p < 0.001) of EF sites.
Using these empirical relationships, the GR model can then be
calibrated and applied to predict GPP. To show the potential of
the calibrated GR model, we applied it to predict GPP for each
Fig. 5. Model calibration using enhanced vegetation index (EVI) and the land surface temperature (LST) for (a and b) non-forest, (c and d) deciduous forest and (e and f)
evergreen forest sites. The Slope and Intercept are the regression coefficients in the relationship between flux GPP and EVI PAR. dEVI, EVI_ave and EVI_sd represent the
difference between maximum and minimum EVI, the average value of EVI, and the standard deviation of EVI, respectively. LST_ave and LST_sd are the average value and the
standard deviation of LST, respectively. Points represent data of each site. Solid and dash lines represent regression and 95% confidence, respectively.
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
PFT and included the MODIS GPP for comparison (Fig. 6). MODIS
GPP generally showed moderate performances in GPP estimations
for the three PFTs with R2 of 0.42 (p < 0.001), 0.42 (p < 0.001) and
0.52 (p < 0.001) for NF, DF and NF sites, respectively. The root mean
square error (RMSE) of estimated GPP ranged from 16.9 g C m2 8 d1 for EF sites to 27.4 m2 8 d1 for DF sites. Furthermore, MODIS
GPP was observed to be higher than flux GPP at the low end of the
range while lower than flux GPP at the upper end of the GPP range
for all PFTs (note the extent may differ among PFTs). This pattern
has been reported in previous analysis of both site level (Harris
and Dash, 2010; Wu et al., 2011) and regional evaluations
(Sjöström et al., 2011). Improper characterization of shaded leaves
in dense canopies by the MODIS GPP algorithm was identified as
the main reason for this pattern (Cheng et al., 2010; Wu et al.,
2012; Zhang et al., 2012; Chen et al., 2012). Using the calibrated
GR model, we observed improved GPP estimates for all PFTs with
all R2 above 0.60 (p < 0.001). DF sites performed the best with R2
of 0.72 (p < 0.001) and a 30% improvement in the RMSE with a value of 19.1 m2 8 d1 compared to MODIS GPP product. The slopes
75
for all of the regressions were close to unity, indicating the robustness of the model calibration approach.
We also compared the calibration results with standard MODIS
GPP product at each site and the slopes, intercepts and R2 for both
models were calculated (not shown here for brevity). Significant
correlations (p < 0.001) were found between GR modeled GPP
and flux-measured GPP for all sites. For MODIS GPP, three sites
have not shown statistically significant correlations with flux-measured GPP (IT-RO2, IT-LEC and IT-BON). GR model produced comparable results with MODIS GPP product for NF and EF sites. The
GR model performed better in most of DF sites compared to standard MODIS GPP product. We further calculated the average slopes
and intercepts as well as their standard deviations for each PFT
(Fig. 7). The average slopes (standard deviation) of MODIS GPP
were 1.29 (0.52), 1.41 (0.46) and 0.92 (0.23) for NF, DF and EF sites,
respectively, implying MODIS GPP would have underestimated results for ND and DF sites. But for EF sites, such underestimation
was not that evident, probably due to a lower range of GPP for
EF sites. The average (standard deviation) slopes of the calibrated
Fig. 6. MODIS 8-day gross primary production (GPP) in (a) non-forest (s), (b) deciduous forest (4), and (c) evergreen forest (h) sites. Modeled GPP using the greenness and
radiation model (GR) in (d) non-forest (s), (e) deciduous forest (4), and (f) evergreen forest (h) sites. Solid and dash lines represent the regression and 1:1 lines, respectively.
76
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
Fig. 7. Comparison between the average slopes and intercepts from individual sites
for (a) MODIS GPP and (b) GR model estimated GPP in three plant functional types.
NF, DF and EF represent non-forest deciduous forest, and evergreen forest
ecosystems, respectively. The error bar indicates the standard deviation (sd) of
each variable for all sites of the same plant functional type and the dash lines
indicate slopes of 1.
GR model for the three PFTs were 1.00 (0.26), 1.00 (0.14) and 0.97
(0.19), indicating improved GPP estimates and model stability.
3.3. Latitudinal patterns of model performance
Grouping vegetation types in different ecoregions may add
complexity as similar vegetation type may be growing under different environmental conditions. We therefore explored the model
performance (i.e., R2 between GR modeled and flux-measured GPP)
as a function of latitudinal gradient which would show how
regional climates (temperature and rain) affect the GR model
performance.
Since there were only three sites in the Southern Hemisphere,
here we only focused on the Northern sites (Fig. 8). We found significant correlations between model performance and latitude
with R2 of 0.38 (p = 0.002) and 0.25 (p = 0.015) for NF and EF sites,
respectively. This indicates that GR model can generally provide
better GPP estimates for sites located at higher latitudes for these
two PFTs. In comparison, DF sites did not show this pattern, suggesting that the GR model is more applicable for DF sites and model performance may be independent on regional climate. These
show that we may expect a combined effect of both PFT and regional climate on the model performance.
Fig. 8. Latitudinal pattern of mode performance as indicated by the coefficients of
determination (R2) between flux-measured GPP and GR modeled GPP for (a) nonforest sites, (b) deciduous forest sites, and (c) evergreen forest sites. Solid and dash
lines represent regression and 95% confidence, respectively.
4. Discussion
4.1. The reason of using radiation to capture short-term GPP variations
The GR model showed promising results for estimating 8-day
GPP for different PFTs. Canopy greenness (i.e., EVI) and radiation
act as two factors regulating the model performance. A single
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
77
vegetation index (e.g., EVI) only provides a measure of canopy
greenness and it cannot capture the influences of short-term
meteorological variables on GPP, especially the variations in temperature. This is the main reason limiting the use of a single index for estimating short-term GPP. Previous analysis by
Sakamoto et al. (2011) demonstrated that daily GPP cannot be
explained by a single WDRVI index but a combination with radiation capturing certain short-term meteorological variations
would help to interpret GPP. To explain this, we explored the
relationship between gridded radiation and the flux-measured
air temperature for all observations (Fig. 9). We found that the
air temperatures of all PFTs were highly correlated with the gridded radiation, indicating that the requirement of independent
temperature may not be necessary for the GR model. This probably is the reason for the wide applicability of the GR model for
such diverse ecosystems in our analysis, especially in capturing
the 8-day GPP variations. We also suggest that it is this potential
that explains the results of GPP estimation using the GR model
even at the daily step (Peng and Gitelson, 2012). One however
should note that our analysis may only support the potential of
radiation in capturing temperature based stress, which may be
not true for water (e.g., soil moisture) stress (e.g., low precision
at US-VAR).
4.2. Physiologically-based interpretation of calibration process
Our previous evaluation of the GR model to estimate monthly
GPP in the growing season of several North America vegetation
ecosystems indicated good performances (Wu et al., 2011). However, the underestimation of high GPP and the overestimation
low GPP was also shown to be the main limitation of the GR model.
The measured GPP from several non-growing season observations
is very low (close to zero) while the GR model considerably overestimated the GPP, indicating both regression slopes and intercepts
should be calibrated carefully. For NF and DF sites, the regression
slopes were correlated with EVI-related parameters, implying that
the dynamical ranges of canopy greenness were the main factor
driving the model performances. By comparison, for EF sites, the
variation in canopy greenness is very small and therefore it may
not be that important. In such situations, the temperature may become an important factor in regulating GPP. Accordingly, we found
no correlation between EVI-related parameters and the regression
slopes while single LSTave can explain as much as 37% variances
(p < 0.001, data not shown here for brevity). Therefore, an integrated form of dEVI LSTave has been found to have the best
potential in explaining slopes of EF sites. For all PFTs, determinations of intercepts were achieved through the inclusion of MODIS
LST. The underlying mechanism probably is that cold temperature
mainly suppresses GPP to zero while the algorithm (EVI PAR)
still gives estimates above zero since there is abundant PAR in most
period of the non-growing season and EF have relatively consistent
EVI values throughout the year. Therefore, inclusion of MODIS LST
would help to remedy this difference and thus to calibrate the
model.
The interesting aspect of our analysis is the physiological interpretation of slopes and intercepts. According to the Monteith logic,
GPP is linearly correlated with LUE and APAR. Since the GR model
is based on the relationships between total chlorophyll content
and GPP, it does not belong to the LUE model. Therefore, the slope
here may not be interpreted as directly equivalent to LUE value in
LUE-based GPP model. The intercept may represent the overestimation or underestimation of GPP using the GR model since it is
obtained when x-axis (i.e., GR model prediction) is zero, showing
the bias compared to the corresponding flux measured GPP value
on the y-axis.
Fig. 9. Relationships between flux-measured air temperature and global radiation
acquired from the National Center for Environmental Prediction (NCEP) reanalysis II
data. The NF (s), DF (4), and EF (h) represent non-forest, deciduous forest, and
evergreen forest ecosystems, respectively.
5. Conclusions
Using 70 global flux sites, we have demonstrated the potential
of a simple remote sensing based GR model for predicting 8-day
GPP across diverse vegetation ecosystems globally. This model is
entirely driven by remote sensing observations (i.e., EVI and LST)
and global coarse resolution radiation data. Model calibration
was achieved by empirical correlations of statistical parameters
78
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
derived from EVI and LST. The calibrated model showed improved
GPP estimates compared to the standard MODIS GPP product.
Previous analysis of this model at monthly time step may only
be an indicator of plant phenology and potentially cannot capture
short-term GPP variations due to regional climate variations (Wu
et al., 2011). However, with the current evaluation, it is suggested
that the GR model can capture short-term GPP variations through
the combination of vegetation index and radiation data. Such
results are useful in developing new GPP algorithms that are entirely based on easily available remote sensing and coarse resolution meteorological observations to circumvent the difficulties in
acquiring ground data globally. Our evaluation also supports the
great potential of the MODIS land surface products in monitoring
ecosystem processes (Ryu et al., 2011). A further investigation will
include mapping the spatial patterns of regional GPP, which would
provide great help for the estimation of global C sequestration.
Acknowledgments
This work was funded by the Open Research Fund of Key Laboratory of Digital Earth Science, Institute of Remote Sensing and
Digital Earth, Chinese Academy of Sciences (2013LDE003), the National Natural Science Foundation of China (Grant Nos. 41001210,
41371013, 41271412), the Knowledge Innovation Program of
Chinese Academy of Sciences (KZCX2-EWQN302) and by Key laboratory funds (KYQ1202). The flux data was acquired by the FLUXNET community and in particular by the following net-works:
AmeriFlux, AfriFlux, AsiaFlux, CarboAfrica, CarboEuropeIP, CarboItaly, CarboMont, ChinaFlux, Fluxnet-Canada, GreenGrass, KoFlux,
LBA, NECC, OzFlux, TCOS-Siberia, and USCCC. We acknowledge
the financial support to the eddy covariance data harmonization
provided by CarboEuropeIP, FAO-GTOS-TCO, iLEAPS, Max Planck
Institute for Biogeochemistry, National Science Foundation,
University of Tuscia, Université Laval and Environment Canada
and US Department of Energy and the database development and
technical support from Berkeley Water Center, Lawrence Berkeley
National Laboratory, Microsoft Research eScience, Oak Ridge
National Laboratory, University of California-Berkeley, University
of Virginia.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.isprsjprs.2013.
10.015.
References
Barr, A.G., Black, T.A., Hogg, E.H., Kljun, N., Morgenstern, K., Nesic, Z., 2004.
Interannual variability in the leaf area index of a boreal aspen-hazelnut forest in
relation to net ecosystem production. Agric. For. Meteorol. 126 (3–4), 237–255.
Beer, C., Reichstein, M., Tomelleri, E., Ciais, P., Jung, M., Carvalhais, N., Rödenbeck, C.,
Arain, M.A., Baldocchi, D., Bonan, G.B., Bondeau, A., Cescatti, A., Lasslop, G.,
Lindroth, A., Lomas, M., Luyssaert, S., Margolis, H., Oleson, K.W., Roupsard, O.,
Veenendaal, E., Viovy, N., Williams, C., Ian Woodward, F., Papale, D., 2010.
Terrestrial gross carbon dioxide uptake: global distribution and covariation
with climate. Science 329 (5993), 834–838.
Betts, A.K., Zhao, M., Dirmeyer, P.A., Beljaars, A.C.M., 2006. Comparison of ERA40
and NCEP/DOE near-surface data sets with other ISLSCP-II data sets. J. Geophys.
Res. 111, D22S04. http://dx.doi.org/10.1029/2006JD007174.
Chen, J.M., Mo, G., Pisek, J., Liu, J., Deng, F., Ishizawa, M., Chan, D., 2012. Effects of
foliage clumping on the estimation of global terrestrial gross primary
productivity. Global Biogeochem. Cycles 26, GB1019. http://dx.doi.org/
10.1029/2010GB003996.
Cheng, Y.B., Middleton, E.M., Huemmrich, K.F., Zhang, Q., Campbell, P.K.E., Corp, L.A.,
Russ, A.L., Kustas, W.P., 2010. Utilizing in situ directional hyperspectral
measurements to validate bio-indicator simulations for a corn crop canopy.
Ecol. Inform. 5 (5), 330–338.
Coops, N.C., Black, T.A., Jassal, R.S., Trofymow, J.A., Morgenstern, K., 2007.
Comparison of MODIS, eddy covariance determined and physiologically
modeled gross primary production (GPP) in a Douglas-fir forest stand.
Remote Sens. Environ. 107 (3), 385–401.
Dash, J., Curran, P.J., 2004. The MERIS terrestrial chlorophyll index. Int. J. Remote
Sens. 25 (23), 5403–5413.
Desai, A.R., Richardson, A.D., Moffat, A.M., Kattge, J., Hollinger, D.Y., Barr, A., Falge, E.,
Noormets, A., Papale, D., Reichstein, M., Stauch, V.J., 2008. Cross site evaluation
of eddy covariance GPP and RE decomposition techniques. Agric. For. Meteorol.
148 (6–7), 821–838.
Field, C.B., Behrenfeld, M.J., Randerson, J.T., Falkowski, P., 1998. Primary production
of the biosphere: integrating terrestrial and oceanic components. Science 281
(5374), 237–240.
Gamon, J.A., Serrano, L., Surfus, J.S., 1997. The photochemical reflectance index: an
optical indicator of photosynthesis radiation use efficiency across species,
functional types, and nutrient levels. Oecologia 112 (4), 492–501.
Gitelson, A.A., 2004. Wide dynamic range vegetation index for remote
quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161
(2), 165–173.
Gitelson, A.A., Viña, A., Masek, J.G., Verma, S.B., Suyker, A.E., 2008. Synoptic
monitoring of gross primary productivity of maize using Landsat data. IEEE
Geosci. Remote Sens. Lett. http://dx.doi.org/10.1109/LGRS.2008.915598.
Gitelson, A.A., Viña, A., Verma, S.B., Rundquist, D.C., Arkebauer, T.J., Keydan, G.,
Leavitt, B., Ciganda, V., Burba, G.G., Suyker, A.E., 2006. Relationship between
gross primary production and chlorophyll content in crops: implications for the
synoptic monitoring of vegetation productivity. J. Geophys. Res. 111, D08S11.
http://dx.doi.org/10.1029/2005JD006017.
Hall, F.G., Hilker, T., Coops, N.C., 2011. PHOTOSYNSAT, photosynthesis from space:
theoretical foundations of a satellite concept and validation from tower and
spaceborne data. Remote Sens. Environ. 115 (8), 1918–1925.
Harris, A., Dash, J., 2010. The potential of the MERIS terrestrial chlorophyll index for
carbon flux estimation. Remote Sens. Environ. 114 (8), 1856–1862.
Hilker, T., Coops, N.C., Hall, F.G., Black, T.A., Wulder, M.A., Nesic, Z., Krishnan, P.,
2008a. Separating physiologically and directionally induced changes in PRI
using BRDF models. Remote Sens. Environ. 112 (6), 2777–2788.
Hilker, T., Coops, N.C., Wulder, M.A., Black, T.A., Guy, R.D., 2008b. The use of remote
sensing in light use efficiency based models of gross primary production: a
review of current status and future requirements. Sci. Total Environ. 404 (2–3),
411–423.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview
of the radiometric and biophysical performance of the MODIS vegetation
indices. Remote Sens. Environ. 83 (1–2), 195–213.
Kalfas, J.L., Xiao, X., Vanegas, D.X., Verma, S.B., Suyker, A.E., 2011. Modeling gross
primary production of irrigated and rain-fed maize using MODIS imagery and
CO2 flux tower data. Agric. For. Meteorol. 151 (12), 1514–1528.
Kanniah, K.D., Beringer, J., Hutley, L.B., Tapper, N.J., Zhu, X., 2009. Evaluation of
collections 4 and 5 of the MODIS gross primary productivity product and
algorithm improvement at a tropical savanna site in northern Australia. Remote
Sens. Environ. 113 (9), 1808–1822.
Monteith, J.L., 1972. Solar radiation and production in tropical ecosystems. J. Appl.
Ecol. 9 (3), 747–766.
Mu, Q., Zhao, M., Running, S.W., 2011. Improvements to a MODIS global terrestrial
evapotranspiration algorithm. Remote Sens. Environ. 115 (8), 1781–1800.
Papale, D., Valentini, A., 2003. A new assessment of European forests carbon
exchange by eddy fluxes and artificial neural network spatialization. Glob.
Change Biol. 9 (4), 525–535.
Papale, D., Reichstein, M., Aubinet, M., Canfora, E., Bernhofer, C., Kutsch, W.,
Longdoz, B., Rambal, S., Valentini, R., Vesala, T., Yakir, D., 2006. Towards a
standardized processing of net ecosystem exchange measured with eddy
covariance technique: algorithms and uncertainty estimation. Biogeosciences 3
(4), 571–583.
Peng, Y., Gitelson, A.A., 2012. Remote estimation of gross primary productivity in
soybean and maize based on total crop chlorophyll content. Remote Sens.
Environ. 117, 440–448.
Peng, Y., Gitelson, A.A., 2011. Application of chlorophyll-related vegetation indices
for remote estimation of maize productivity. Agric. For. Meteorol. 151 (9),
1267–1276.
Peng, Y., Gitelson, A.A., Keydan, G., Rundquist, D.C., Moses, W., 2011. Remote
estimation of gross primary production in maize and support for a new
paradigm based on total crop chlorophyll content. Remote Sens. Environ. 115
(4), 978–989.
Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M., Berbigier, P.,
Bernhofer, C., Buchmann, N., Gilmanov, T., Granier, A., Grunwald, T.,
Havrankova, K., Ilvesniemi, H., Janous, D., Knohl, A., Laurila, T., Lohila, A.,
Loustau, D., Matteucci, G., Meyers, T., Miglietta, F., Ourcival, J.M., Pumpanen, J.,
Rambal, S., Rotenberg, E., Sanz, M., Tenhunen, J., Seufert, G., Vaccari, F., Vesala,
T., Yakir, D., Valentini, R., 2005. On the separation of net ecosystem exchange
into assimilation and ecosystem respiration: review and improved algorithm.
Glob. Change Biol. 11 (9), 1424–1439.
Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., Harlan, J.C., 1974. Monitoring the
vernal advancements and retrogradation of natural vegetation. In: NASA/GSFC.
Final Report. Greenbelt, MD, USA, pp. 1–137.
Running, S.W., Nemani, R.R., Heinsch, F.A., Zhao, M., Reeves, M., Hashimoto, H.,
2004. A continuous satellite-derived measure of global terrestrial primary
production. Bioscience 54 (6), 547–560.
Ryu, Y., Baldocchi, D.D., Kobayashi, H., van Ingen, C., Li, J., Black, T.A., Beringer, J., van
Gorsel, E., Knohl, A., Law, B.E., Roupsard, O., 2011. Integration of MODIS land and
atmosphere products with a coupled-process model to estimate gross primary
C. Wu et al. / ISPRS Journal of Photogrammetry and Remote Sensing 88 (2014) 69–79
productivity and evapotranspiration from 1 km to global scales. Global
Biogeochemical Cycles, 25(GB4017).http://dx.doi.org/10.1029/2011GB004053.
Sakamoto, T., Gitelson, A.A., Wardlow, B.D., Verma, S.B., Suyker, A.E., 2011.
Estimating daily gross primary production of maize based only on MODIS
WDRVI and shortwave radiation data. Remote Sens. Environ. 115 (12), 3091–
3101.
Sims, D.A., Rahman, A.F., Cordova, V.D., El-Masri, B.Z., Baldocchi, D.D., Bolstad, P.V.,
Flanagan, L.B., Goldstein, A.H., Hollinger, D.Y., Misson, L., Monson, R.K., Oechel,
W.C., Schmid, H.P., Wofsy, S.C., Xu, L., 2008. A new model of gross primary
productivity for North American ecosystems based solely on the enhanced
vegetation index and land surface temperature from MODIS. Remote Sens.
Environ. 112 (4), 1633–1646.
Sjöström, M., Ardö, J., Arneth, A., Boulain, N., Cappelaere, B., Eklundh, L., de
Grandcourt, A., Kutsch, W.L., Merbold, L., Nouvellon, Y., Scholes, R.J., Schubert,
P., Seaquist, J., Veenendaal, E.M., 2011. Exploring the potential of MODIS EVI for
modeling gross primary production across African ecosystems. Remote Sens.
Environ. 115 (4), 1081–1089.
Turner, D.P., Urbanski, S., Bremer, D., Wofsy, S.C., Meyers, T., Gower, S.T., Gregory,
M., 2003. A cross-biome comparison of daily light use efficiency for gross
primary production. Glob. Change Biol. 9 (3), 383–395.
Vermote, E.F., El Saleous, N., Justice, C.O., Kaufman, Y.J., Privette, J.L., Remer, L.,
Roger, J.C., Tanré, D., 1997. Atmospheric correction of visible to middle-infrared
EOS-MODIS data over land surfaces: background, operational algorithm and
validation. J. Geophys. Res. 102 (D14), 17131–17141.
Wan, Z., 2008. New refinements and validation of the MODIS land-surface
temperature/emissivity products. Remote Sens. Environ. 112 (1), 59–74.
Wang, H., Jia, G., Fu, C., Feng, J., Zhao, T., Ma, Z., 2010. Deriving maximal light use
efficiency from coordinated flux measurements and satellite data for regional
gross primary production modeling. Remote Sens. Environ. 114 (10), 2248–
2258.
Wu, C., Chen, J.M., Huang, N., 2011. Predicting gross primary production from the
enhanced vegetation index and photosynthetically active radiation: evaluation
and calibration. Remote Sens. Environ. 115 (12), 3424–3435.
Wu, C., Chen, J.M., Desai, A.R., Hollinger, D.Y., Arain, M.A., Margolis, H.A., Gough,
C.M., Staebler, R.M., 2012. Remote sensing of canopy light use efficiency in
79
temperate and boreal forests of North America using MODIS imagery. Remote
Sens. Environ. 118, 60–72.
Wu, C., Munger, J.W., Niu, Z., Kuang, D., 2010. Comparison of multiple models for
estimating gross primary production using MODIS and eddy covariance data in
Harvard Forest. Remote Sens. Environ. 114 (12), 2925–2939.
Wu, C., Niu, Z., Tang, Q., Huang, W., Rivard, B., Feng, J., 2009. Remote estimation of
gross primary production in wheat using chlorophyll-related vegetation
indices. Agric. For. Meteorol. 149 (6–7), 1015–1021.
Xiao, X.M., Boles, S., Liu, J.Y., Zhuang, D.F., Frolking, S., Li, C.S., Salas, W., Moore, B.,
2005. Mapping paddy rice agriculture in southern China using multi-temporal
MODIS images. Remote Sens. Environ. 95 (4), 480–492.
Xiao, X., Boles, S., Frolking, S., Li, C., Babu, J.Y., Salas, W., Berrien Moore, I.I.I., 2006.
Mapping paddy rice agriculture in South and Southeast Asia using multitemporal MODIS images. Remote Sens. Environ. 100 (1), 95–113.
Yuan, W., Liu, S., Yu, G., Bonnefond, J., Chen, J., Davis, K., Desai, A.R., Goldstein, A.H.,
Gianelle, D., Rossi, F., Suyker, A.E., Verma, S.B., 2010. Global estimates of
evapotranspiration and gross primary production based on MODIS and global
meteorology data. Remote Sens. Environ. 114 (7), 1416–1431.
Zhang, Q., Middleton, E.M., Margolis, H.A., Drolet, G.G., Barr, A.A., Black, T.A., 2009.
Can a satellite-derived estimate of the fraction of PAR absorbed by chlorophyll
(FAPARchl) improve predictions of light-use efficiency and ecosystem
photosynthesis for a boreal Aspen forest? Remote Sens. Environ. 113 (4),
880–888.
Zhang, F., Chen, J.M., Chen, J., Gough, C.M., Martin, T., Dragoni, D., 2012. Evaluating
spatial and temporal patterns of MODIS GPP over the conterminous U.S. against
flux measurements and a process model. Remote Sens. Environ. 124, 717–729.
Zhao, M., Running, S.W., 2010. Drought-induced reduction in global terrestrial net
primary production from 2000 through 2009. Science 329 (5994), 940–943.
Zhao, M., Heinsch, F.A., Nemani, R.R., Running, S.W., 2005. Improvement of the
MODIS terrestrial gross and net primary production global dataset. Remote
Sens. Environ. 95 (2), 164–176.
Zhao, M., Running, S.W., Nemani, R.R., 2006. Sensitivity of Moderate Resolution
Imaging Spectroradiometer (MODIS) terrestrial primary production to the
accuracy of meteorological reanalyses. J. Geophys. Res. 111, G01002. http://
dx.doi.org/10.1029/2004JG000004.